Rows 2+3 packets: Optional carrier fork + NonEmptyStr renderer residue root-cause (materialization_carriers, no hot-emitter edits)

[gunbai-bot]

Summary

Root-cause-only doc (no repair) for materialization_carriers.dag error-board Rows 2 and 3.
Measured against origin/main a2b33f577a48b13e305440bae41d90406e8b4b7d, re-checked for drift
against current origin/main 073bbd4e08a40a01082c61896a2f2806d683c6c2 (no touch to the relevant
files in between). PR #8528 confirmed OPEN/unmerged at measurement time, so Row 3 carries all 5
of its original sites.

• Row 2 ("Optional carrier fork", 6 E0308 sites) splits into three sub-mechanisms, not one:
  • 2a (std_cache_interface.rs:638) and 2c (extdeps_uri.rs:752,756): a bare zero-field-variant
    identifier (None/Absent) resolves to the wrong parent enum because an unrelated domain
    coproduct in the same closure happens to declare a variant with the same name.
    contextual_variant_parent_absent (src/v1/05_emit_rust.dag) is confirmed correct given its
    inputs; the wrong resolved_type is fed to it from upstream in 04_infer.dag.
  • 2b (std_cache_interface.rs:695,699,703): three functions declare a non-optional return type
    with a body ending in cardinality-optional List.first(); gunbc compile reports 0 blocking
    errors, only rustc catches it — an apparent .dag-level typecheck gap.
• Row 3 residual (3 E0425 NonEmptyStr sites): traced via a decisive differential — the
  same closed brand alias (CacheInterfaceId) renders correctly when a function's body is an
  ordinary expression, but loses its alias identity to the base type (NonEmptyStr) when the
  body is a bare point-free reference to another function. render_rust_fn_sig_type's
  leaf-preservation guard (rust_fn_sig_preserves_authored_alias_leaf) is confirmed correct and
  simply never reached for these three sites — the loss happens in 04_infer.dag's unification of
  a declared return annotation against a point-free body.

Scope boundary — no hot-emitter edits

src/v1/05_emit_rust.dag and src/v1/04_infer.dag were read-only for this investigation and are
untouched by this branch, per explicit session scope: both files are under heavy concurrent
contention from other in-flight sessions (PR #8528, the T7/row-1a/1b lanes). This PR is
documentation only — one new file under docs/probes/.

Test plan

• Doc-only change; no build/test surface touched.
• All site counts and line numbers re-measured against current origin/main via
  docs/probes/curated_cargo_probe_one.sh, not transcribed from prior docs.
• gh pr view 8528 --json state,mergedAt checked directly (OPEN, mergedAt: null).

🤖 Generated with Claude Code